Has order of an attributes impact on neural network learnig? I want to ask if order of attributes can impact on neural network learning. Is it the same if the  input vector looks like this: 
a b c d e

and like this:
e a d b c

Thank you for your answers.
 A: It depends on the network.
In a convolutional neural network, or any sort of network without full connectivity between layers, it will matter. But in a standard fully connected network it will make no difference at all, even when doing image classification. 
The reasoning behind this is that the output of a neuron is the sum of its inputs multiplied by the weight connecting them. If you change the order of the numbers in a sum you still get the same answer. 
Now what WILL matter is consistancy. If you have the order of your inputs as "a c d b e" for one training example, it will have to be the same for all other training data.
A: I would like to highlight that the accepted answer is not entirely true.
This may be the case after the input layer, but if you use a linear layer as your input layer order does indeed matter. This is because the first linear layers weight is directly related to the order of the input features.
This is because the dot products between the weight matrix and the input matrix will change depending on the order of the input matrix.
This is especially true for 1-D time series data
